File size: 8,163 Bytes
1dbac36 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 | #!/usr/bin/env python3
"""
EA Baselines Comparison - All baselines with and without EvoRM
Uses ref_pairs order for entity selection.
"""
import os, sys, json, time, shutil
from datetime import datetime
sys.path.insert(0, '/root/autodl-tmp/AdaCoAgentEA')
sys.path.insert(0, '/root/autodl-tmp/AdaCoAgentEA/Area2')
sys.path.insert(0, '/root/autodl-tmp/AdaCoAgentEA/baselines/evorm_wrappers')
os.environ["OPENAI_API_BASE"] = "https://hk.xty.app/v1"
os.environ["OPENAI_API_KEY"] = "sk-7a7Ev4VcVyysPLT5hqtqIVD6PybzJ1ZlEIVZddIR3NtZvPgK"
DATA_DIR = "/root/autodl-tmp/AdaCoAgentEA/data/icews_wiki"
RESULTS_DIR = "/root/autodl-tmp/AdaCoAgentEA/results"
ALIGNED_FILE = os.path.join(DATA_DIR, "message_pool", "aligned_entities.txt")
REF_PAIRS_FILE = os.path.join(DATA_DIR, "ref_pairs")
IMPORTANT_FILE = os.path.join(DATA_DIR, "message_pool", "important_entities.txt")
RETRIEVER_FILE = os.path.join(DATA_DIR, "message_pool", "retriever_outputs.txt")
EVORM_STATE = os.path.join(DATA_DIR, "evorm_state")
os.makedirs(RESULTS_DIR, exist_ok=True)
def load_ref_pairs():
pairs = {}
with open(REF_PAIRS_FILE, 'r') as f:
for line in f:
parts = line.strip().split('\t')
if len(parts) == 2:
try:
pairs[int(parts[0])] = int(parts[1])
except:
pass
return pairs
def compute_metrics(ref_subset=None):
aligned = set()
if os.path.exists(ALIGNED_FILE):
with open(ALIGNED_FILE, 'r') as f:
for line in f:
parts = line.strip().split('\t')
if len(parts) == 2:
try:
aligned.add((int(parts[0]), int(parts[1])))
except:
continue
if ref_subset:
ref_set = set(ref_subset.items())
else:
ref_set = set(load_ref_pairs().items())
if not ref_set:
return {'error': 'No ref pairs'}
tp = len(aligned & ref_set)
fp = len(aligned - ref_set)
fn = len(ref_set - aligned)
return {
'Hits@1': round(tp / max(1, len(ref_set)), 4),
'MRR': round(tp / max(1, len(ref_set)), 4),
'Precision': round(tp / max(1, tp + fp), 4),
'Recall': round(tp / max(1, tp + fn), 4),
'F1': round(2 * tp / max(1, 2*tp + fp + fn), 4),
'TP': tp, 'FP': fp, 'FN': fn,
'Total_Ref': len(ref_set),
'Total_Aligned': len(aligned),
}
def clear_state():
if os.path.exists(EVORM_STATE):
shutil.rmtree(EVORM_STATE)
if os.path.exists(ALIGNED_FILE):
os.remove(ALIGNED_FILE)
def setup_subset(n_entities, ref_map):
"""Create subset using ref_pairs order."""
ref_order = []
with open(REF_PAIRS_FILE, 'r') as f:
for line in f:
parts = line.strip().split('\t')
if len(parts) == 2:
try:
e1 = int(parts[0])
if e1 not in ref_order:
ref_order.append(e1)
except:
pass
actual_n = min(n_entities, len(ref_order))
subset_kg1 = set(ref_order[:actual_n])
subset_lines = []
with open(RETRIEVER_FILE, 'r') as f:
for line in f:
parts = line.strip().split('\t')
if len(parts) == 2:
try:
if int(parts[0]) in subset_kg1:
subset_lines.append(line)
except:
pass
with open(IMPORTANT_FILE, 'w') as f:
f.writelines(subset_lines)
subset_ref = {e1: ref_map[e1] for e1 in subset_kg1 if e1 in ref_map}
return subset_ref
def run_adacoagent_baseline(data_dir, use_evorm=False):
"""Run AdaCoAgentEA (baseline or +EvoRM)."""
clear_state()
import tokens_cal
tokens_cal.global_tokens = 0
t0 = time.time()
if use_evorm:
from LLM1_label_selector import align_entities
else:
from LLM1_label_selector_baseline import align_entities
aligned = align_entities(data_dir, from_m3=False)
elapsed = time.time() - t0
return aligned, elapsed, tokens_cal.global_tokens
def main():
n_entities = 200 # Subset size for comparison
ref_map = load_ref_pairs()
print(f"Loaded {len(ref_map)} reference pairs")
subset_ref = setup_subset(n_entities, ref_map)
print(f"Subset: {len(subset_ref)} entities with ref pairs")
# Baselines to run
baselines = [
('AdaCoAgentEA', 'adacoagent'),
('ZeroCoT', 'zerocot'),
('Self-Consistency', 'self_consistency'),
('ChatEA', 'chatea'),
('Collaboration-Hard', 'cohard'),
]
import tokens_cal
results = []
for bl_name, bl_key in baselines:
for use_evorm in [False, True]:
mode = f"{bl_name}{' + EvoRM' if use_evorm else ''}"
print(f"\n{'='*50}")
print(f" {mode}")
print(f"{'='*50}")
clear_state()
setup_subset(n_entities, ref_map) # Re-setup after clear_state
try:
tokens_cal.global_tokens = 0
t0 = time.time()
if bl_key == 'adacoagent':
if use_evorm:
from LLM1_label_selector import align_entities
else:
from LLM1_label_selector_baseline import align_entities
aligned = align_entities(DATA_DIR, from_m3=False)
elif bl_key == 'zerocot':
from zero_cot import align_entities as zc
aligned = zc(DATA_DIR, use_evorm=use_evorm)
elif bl_key == 'self_consistency':
from self_consistency import align_entities as sc
aligned = sc(DATA_DIR, use_evorm=use_evorm)
elif bl_key == 'chatea':
from chat_ea import align_entities as ce
aligned = ce(DATA_DIR, use_evorm=use_evorm)
elif bl_key == 'cohard':
from cohard import align_entities as ch
aligned = ch(DATA_DIR, use_evorm=use_evorm)
else:
raise ValueError(f"Unknown: {bl_key}")
elapsed = time.time() - t0
tokens = tokens_cal.global_tokens
metrics = compute_metrics(ref_subset=subset_ref)
result = {
'method': mode,
'aligned_pairs': len(aligned),
'time_seconds': round(elapsed, 1),
'tokens': tokens,
'tokens_per_pair': round(tokens / max(1, len(aligned)), 1),
'metrics': metrics,
}
results.append(result)
print(f" Aligned: {len(aligned)}, Time: {elapsed:.1f}s, Tokens: {tokens}")
print(f" Hits@1: {metrics['Hits@1']}, Precision: {metrics['Precision']}, Recall: {metrics['Recall']}")
except Exception as e:
print(f" FAILED: {e}")
import traceback
traceback.print_exc()
# Save
timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
result_file = os.path.join(RESULTS_DIR, f"baselines_comparison_{timestamp}.json")
with open(result_file, 'w') as f:
json.dump({
'experiment': 'EA Baselines Comparison',
'dataset': 'ICEWS-WIKI',
'n_entities': n_entities,
'results': results,
}, f, indent=2)
# Summary
print(f"\n{'='*80}")
print("BASELINES COMPARISON SUMMARY")
print(f"{'='*80}")
print(f"{'Method':<35} {'Hits@1':>8} {'Prec':>8} {'Recall':>8} {'F1':>8} {'Tokens':>10} {'Time':>8}")
print("-" * 80)
for r in results:
m = r['metrics']
print(f"{r['method']:<35} {m['Hits@1']:>8.4f} {m['Precision']:>8.4f} "
f"{m['Recall']:>8.4f} {m['F1']:>8.4f} {r['tokens']:>10} {r['time_seconds']:>7.1f}s")
print(f"\nResults: {result_file}")
if __name__ == '__main__':
main()
|